Aktuelle Publikationen

Network canonization has recently been introduced, restructuring a neural network model into a functionally identical equivalent to which established explanation methods can be applied optimally. In this work, we quantitatively verify the...

This paper presents an improved probability estimation scheme for the entropy coder of Incremental Neural Network Coding (INNC), which is currently under standardization in ISO/IEC MPEG. Major finding is that the probability of a significant...

In this work, we aim to understand what type of pretraining works reliably in practice and what type of pretraining dataset is best suited to achieve good performance in small target dataset size scenarios. Considering diabetic retinopathy...

In the evolution of optical networks, enhancement of spectral efficiency (SE) enhancement has been the most cost-efficient and thus the main driver for capacity enhancementincrease for decades. As a result, the development of optical transport...

This work proposes a parallel neural network based demosaicing procedure for single-camera-one-shot-for hyperspectral imaging trained on a new ground truth dataset captured in a controlled environment by a hyperspectral snapshot camera with a 4×4...

We experimentally demonstrate tap-proof coherent optical 640-Gb/s transmission based on encryption-less physical layer security. Information bit scrambling combined with soft-decision error-correction coding yields favorably small security gaps,...

We study the benefits of applying advanced DSP-based monitoring on multiple wavelength division multiplexing (WDM) channels allocated in the optical grid to infer wavelength-wise characteristics of a C+L-band optical line system. In that context,...

We present different concepts of a THz-wireless point-to-multipoint (P2MP) communication system based on fast spatio-temporal beam-switching in the THz domain and discuss implications for the system and the DSP design. Characterization results...

We present an approach to assess the perfusion of visible human tissue from RGB video files. We show that locally resolved rPPG-signals can be used for intraoperative perfusion analysis and visualization during skin and organ transplantation as...

The ability to continuously process and retain new information like we do naturally as humans is a feat that is highly sought after when training neural networks. Unfortunately, the traditional optimization algorithms often require large amounts...